Integrate Compute for GPU Nodes into an Existing System Architecture POWEREDGE-XE-OPERATE Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   When determining hardware compatibility for a GPU node, which specification must be checked to ensure the GPU can be physically installed in the server?

   • AThe GPU's CUDA core count
   • BThe GPU's memory clock speed
   • CThe PCIe slot type and physical dimensions of the GPU
   • DThe GPU's driver version
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A company is planning to add GPU nodes to its existing HPC cluster over the next three years. The initial deployment will be four nodes, but the company expects to scale to 32 nodes. The current cluster uses 25 GbE networking and has available rack space and power. The company has a limited budget and wants to minimize upfront costs while ensuring the architecture can scale. What is the best approach?

   • AUse the existing 25 GbE core switches and add more ports as nodes are added
   • BPurchase all 32 nodes upfront to avoid future compatibility issues
   • CDeploy a leaf-spine network with 100 GbE spine links and provision power/cooling for the full 32-node deployment
   • DDeploy the four nodes with the existing 25 GbE network and upgrade to a leaf-spine topology when the node count reaches 16
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   Which monitoring metric is most important to track for GPU nodes to prevent overheating and potential hardware failure?

   • ADisk I/O throughput
   • BNetwork packet loss
   • CGPU temperature
   • DMemory usage
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A company is planning to add GPU nodes to its existing HPC cluster over the next two years. The initial deployment will be two nodes, but the company expects to scale to 16 nodes. The current cluster uses 25 GbE networking and has available rack space and power. What should the administrator consider to ensure the architecture can scale without a major redesign?

   • AUse the existing 25 GbE core switches and add more ports as nodes are added
   • BPurchase all 16 nodes upfront to avoid future compatibility issues
   • CDeploy the two nodes with the minimum required power and cooling, and upgrade later as needed
   • DDesign the network with a leaf-spine topology and provision power/cooling capacity for the full 16-node deployment
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A GPU node in a production cluster is experiencing intermittent GPU errors in the system log. The errors mention 'Xid 79' and the GPU is resetting. The administrator has already updated the NVIDIA driver to the latest version. What should the administrator do next?

   • AReplace the GPU immediately
   • BReboot the node and ignore the errors if they do not recur
   • CCheck the GPU's power and thermal status via nvidia-smi and review the node's hardware logs
   • DDisable the GPU's ECC to prevent the errors
   Reveal Solution Discussion Select an answer first